The Learning Innovation Incubator at Vanderbilt

Modern AI systems like ChatGPT or Gemini rely on Large Language Models (LLMs) “foundation models” that process data (words, pixels, audio spectra), treating each observation in isolation of the people behind it - an ecological fallacy of disregarding the individuals and communities that generate data.

 In this talk, Andy will discuss ongoing work toward reconceptualizing the foundational probabilistic tasks of AI to integrate the people behind the data. For LLMs, this means updating their foundational task which estimates the probability of the next word from the preceding words to also account for the words' authors. This “human language modeling” (HuLM) framework explicitly conditions on dynamic user states, drawing on theories of traits and states from psychology, to capture the structured dependencies among data. He will discuss implementing these models and show evidence that they can lead to improved performance on both traditional NLP tasks as well as health and psychological applications. Join us on Oct 1 to learn more.

Submit your proposal by Sept 26.

LIVE Ignite Tech Development Grants provide concentrated support from LIVE’s research engineer to bring innovative learning technology ideas to life or help existing tools take bold new steps forward.

Have you ever wondered how AI models know exactly how to answer your question? Maybe you've wondered why AI gets your questions wrong?

David Tovar offered some answers to these questions as he kicked off this semester's LIVE Learning Innovation Seminar Series. Often AI models are fed large data sets but miss an important human element thus resulting in several errors in response. Tovar's session, “Growing Minds, Growing Machines” discussed how his work seeks to develop personalized, brain-inspired AI models that learn from people directly. They start with how many people judge visual similarity, then add patterns found in brain activity from groups, and finally tune to an individual person’s own signals. The result is an AI that better reflects how a specific person perceives the world. Tovar’s work can help redefine how LLMs are used and create a more human user experience. When models are aligned with real human behavior and brain dynamics, assistants can respond in ways that feel natural, anticipate needs, and adapt to each learner without constant retraining. This work can help advance support strategies by offering study tools and tutoring that fit each person’s strengths, by improving recognition and retrieval for users with different learning profiles, and by informing accommodations that match an individual’s decision patterns.

Tovar was inspired to do this work because he aims to bridge how the brain learns with how machines learn, so that AI not only performs well but also supports people more effectively in classrooms, clinics, and everyday life.

Interested in sharing your work with the LIVE community? Email us at LIVE@vanderbilt.edu

Vanderbilt's Center for Technology Transfer & Commercialization (CTTC) is hosting the first session in their new lunch and learn series with Alan Bently, Assistant Vice Chancellor for Technology Transfer & Intellectual Property Development. In the session Alan will host a discussion on navigating innovation and intellectual property. 

Where: Vanderbilt Alumni Hall; Memorial Room

When: Tues Sept 23, 2025 | 12p - 1p

October 1

Human-Centering LLMs at their Foundation

Andy Schwartz (College of Connected Computing)

October 8

Fall Break

No Session

 

October 15

LLM Persona-Agents for Social Science Research

Lisa Fazio (Peabody College)

Explore the full lineup of Fall 2025 events at the link below